Summary

Mega Tech earnings show that the positive loop of AI investment, returns, and reinvestment remains intact. Meta attributed the renewed acceleration in advertising to AI-driven gains in engagement time and targeting precision; even with Amazon’s weaker outlook, its advertising business remained relatively upbeat, and every company said it would continue investing, with Google adding more than $10B. “Has the loop been interrupted? … Actually, it hasn’t.”
The deterioration in U.S. employment data is real, but the speaker rejects the claim that the Bureau of Labor Statistics is fabricating the numbers. May and June payrolls were revised down to 19,000 and 14,000; low response rates, seasonal adjustments, and distortions in counting undocumented immigrants may all have amplified the volatility, while QCEW data show that payrolls do require a significant adjustment. Goldman Sachs estimates that the figures from April last year through March this year could be revised by 500K–950K. Trump’s removal of the BLS chief was therefore described as “quite nonsensical.”
Roughly $600B in tariffs amounts to an equivalent consumption tax, which the speaker estimates will raise inflation by nearly 1 percentage point while cutting growth by nearly 1 percentage point. The effective tariff rate is expected to be around 18%, but the roughly 15% agreements with Europe, Japan, and South Korea, along with exemptions covering about half of copper imports, suggest that while Trump’s rhetoric is loud, implementation may gradually loosen; pharmaceuticals and chips are also more likely to be delayed or come with extensive exceptions.
In Q3 and Q4 2025, the fiscal timing gap means the negatives arrive first and the positives later, with the worst data potentially extending into Q1 next year. The tariff shock has already begun and will continue to feed through with a lag, while government spending will contribute only a few dozen ticks to this year’s growth. If Trump’s tax cuts are implemented in full, their positive feedback will not appear until 2026. This is a “double-edged sword” that depresses growth while lifting inflation.
U.S. equities are headed for a split-screen market: economically sensitive earnings and advertising will come under pressure, but AI earnings, capex, and token demand may not be derailed by a slowdown. The speaker expects token usage to rise exponentially and sees the supply shortfall as difficult to reverse. Markets can still correct when the economy first turns down, so investors who have not entered can wait for an opportunity and then capture the AI supercycle. “Weak data does not necessarily mean stocks will plunge.”
With payrolls sharply revised down and Powell having characterized tariff inflation as “one-off,” the speaker’s clear call is for a 25-basis-point rate cut in September. Companies are currently absorbing roughly 50%–60% of the tariffs, and while the price shock will feed through gradually, that does not change its one-off nature. A 50-basis-point cut is unlikely because “that would also mean the Fed had made a policy mistake earlier.”

Deep dive

1. Earnings confirm that the AI investment and monetization loop remains in motion

The speaker’s central takeaway from Big Tech earnings was: “Keep investing, earn returns on that investment, then invest even more.” That loop has not been interrupted. AI is no longer just a capex narrative; it is generating verifiable productivity gains in cloud and advertising.
Meta is the clearest example. Advertising, historically more cyclical and not a high-growth business, has reaccelerated, and the company explicitly attributed that to AI: AI is driving longer engagement times and more precise ad targeting, directly accelerating advertising growth.
Even Amazon, one of the weaker companies on guidance, remained relatively optimistic about advertising growth and continued to say it would invest. Microsoft’s related growth had been expected at roughly 39%, potentially moving into the low 40s; the market had feared further deterioration, but that did not happen. Google, meanwhile, increased investment by more than $10B.

2. Payroll revisions reflect a weakening economy, not deliberate official fabrication

Trump removed the BLS chief after the May and June figures were sharply revised down, believing the data looked good under Biden but deteriorated just as he was preparing to declare a policy victory. The speaker’s rebuttal was direct: “The possibility of large-scale fraud is almost nonexistent,” because it would constitute a crime in the United States.
The revisions first reflect statistical mechanics. Education jobs across the states have been heavily affected by policy, while business and household surveys cannot be fully collected in real time. With response rates low this round, the initial estimates had to rely on limited feedback and seasonal adjustments; significant subsequent revisions are therefore not unusual.
Undocumented immigration further weakens the match between the two employment datasets. The people captured in the household survey may not include undocumented immigrants, while the establishment survey may count them. Policies targeting undocumented immigrants can also distort employer reporting when companies declare that workers lack labor visas. As a result, payrolls and the unemployment rate still point in the same direction, including over the past year, but their alignment is weaker than before.
QCEW, which tracks relatively well with payrolls, indicates that payroll employment does require a substantial adjustment. Goldman Sachs estimates that data covering the past year—from April last year through March this year—could be revised by 500K–950K, a larger range than last year. The speaker expects economic data to remain weak through Q3 and Q4, potentially extending into Q1 next year.

3. Tariffs weigh on growth in aggregate, while implementation keeps adding exemptions

The speaker equated roughly $600B in tariffs with “adding $600B in consumption tax.” An effective tariff rate of around 18% could lift inflation by nearly 1 percentage point and reduce growth by nearly 1 percentage point, cutting both ends with the same “double-edged sword.”
Europe, Japan, and South Korea face rates of roughly 15% and together account for about one-third of U.S. imports. Mexico has secured a delay, while Bessent has signaled that China may also receive one. Canada’s headline rate is 35%, but it applies only to goods that do not qualify under USMCA; only the low-teens percentage of Canadian exports are estimated to actually face the 35% tariff.
Copper tariffs offer a template for how implementation may work. Although tariffs were imposed nominally, about half of the products were ultimately exempted. The speaker therefore expects sector-based tariffs to be announced with fanfare and then delayed or accompanied by extensive exceptions.
Tariffs on pharmaceuticals are unpopular and could affect the midterm elections. There is also limited room for broad tariffs on chips. TSMC’s U.S. fabs have not yet expanded at scale or entered mass production, while Intel’s capacity is primarily for its own use; imposing universal chip tariffs would therefore amount to a direct tax on the high-tech sector. Delays or extensive carve-outs are more likely.

4. Tariffs arrive before tax cuts, pushing markets toward a split-screen outcome

The fiscal timing gap is the main risk for Q3 and Q4. Government spending will still have little impact on 2025 growth—only a few dozen ticks, roughly 20 ticks or so—while the positive feedback from Trump’s tax cuts, if fully implemented, will not arrive until 2026. The negative effects of tariffs, however, are already feeding through with a lag.
The speaker expects the market to bifurcate. A weakening economy will inevitably drag down overall corporate earnings and reduce cyclical advertising demand, but it “may not be enough to stop” AI companies’ earnings and investment momentum. Big Tech could continue spending on capex through several quarters of weak data.
The positioning advice is therefore measured. When the economy first turns down, asset prices are typically hit, so investors who have not entered can wait. If an opportunity emerges, they can gain exposure to semiconductors and the AI supercycle: token demand will rise exponentially, and the broader supply shortfall will be difficult to reverse.

5. After employment breached the warning line, a 25-basis-point September cut becomes the base case

Trump’s pressure on Powell is not the key issue. The speaker views the Fed as a professional, data-driven institution. Powell made 3 clear statements last week characterizing tariff-driven inflation as “one-off.” Even if PCE rises to 3.3% rather than remaining below the previous 2.5% forecast, the Fed will prioritize growth when the economy clearly weakens unless inflation becomes extremely severe.
“One-off” does not mean immediate. Companies are currently absorbing roughly 50%–60% of the tariffs, with the costs to be passed through gradually. From December last year through June this year, core services fell from 3.9% to 3.4%, while core goods rose from near zero or negative territory to around 1%, consistent with a tariff-driven goods inflation shock.
If the original May and June payroll figures of 140,000 and 147,000 had held, a rate cut might have been pushed later in the year, potentially to December. But the revised figures of 19,000 and 14,000 are now below the speaker’s 80,000 warning threshold. With growth already weak and inflation characterized by the Fed itself as one-off, “what reason do you have to explain” not cutting rates?
The base case is a 25-basis-point cut in September. Even if the Fed does not cut, it should offer very clear guidance on subsequent easing. A 50-basis-point move is unlikely because it would imply that policy had previously been too slow. Markets have not suffered a decline large enough to warrant it, the reaction has been broadly positive, and the Fed also dislikes excessive asset-market bubbles.

The Data on America's Economic Split | Andrew Milgram Interview 46. So-Young's 金星 in His Own Words: 6 Years as a Nasdaq-Listed Company, from Peak to Bottom, and My Fourth Start